bots don't cry

Autoregressively sewing and generating garments from text guidance

[bots don't cry @botsdontcry1]

@DaveSteinSays You get paid by @Adobe to tell their C2PA narrative.
How do you know that I'm not contributing? There is a probably a reason why I can explain the C2PA fiasco to journalists, researchers: the industry is listening.
I may also be @Adobe or @Google employee, but not C2PA conform.

[bots don't cry @botsdontcry1]

@MikeDoris @talhagin @GeminiApp As for "conformant" C2PA images, it's utterly garbage to put it mildly:
Gemini generated 2025-12-06 14:29:00, with 11 manifests, and a total of 14 actions and 130 ingredients, no "c2pa.filtered" actions.

Releasing something this wrong and misleading is beyond irresponsible.

[bots don't cry @botsdontcry1]

@MikeDoris @talhagin The @GeminiApp accepts only "conformant" C2PA. Everything else rely on CV + synthID.
Only Vivo cameras are conform, outside of OpenAI and Google.
There is probably close to 0 conform C2PA data in Gemini training data, thus the LLM classic "hallucination".
https://t.co/wILoyhhVH0

[bots don't cry @botsdontcry1]

Great! @Google has upgraded its AI image verification.

So let's try it again with a good old DALL·E 3 image.

The verdict? It STILL fails spectacularly.

Worse: it ends up granting legitimacy to the very fakes it’s meant to expose.

[bots don't cry @botsdontcry1]

Claim 5:
"Google ensures maximum privacy"

➡️ Yes, trust Google, much like the Adobe C2PA architect and Google’s own C2PA lead, who publicly upload their Pixel‑generated images to GitHub.https://t.co/ZvFvXhMIf5

Metadata, with Google "Maximum privacy":
https://t.co/kuVpDqtgBn

[bots don't cry @botsdontcry1]

Responses from @Google to questions about C2PA by @KateKayeReports need a proper debunk

Claim 1:
"OpenAI Verify clearly says it's only for verifying whether media was produced by OpenAI tools"

➡️They don't need C2PA conformance for that. But they need it to please regulators.

[bots don't cry @botsdontcry1]

Claim 4:
"No one is forced to use conformant implementation"
➡️ That's what Adobe, Bytedance, Leica, Sony, Nikon do, with their own proprietary tools. No interoperability.

But how Adobe, with non conform, deprecated manifests for most of its software get validated on Linkedin?

[bots don't cry @botsdontcry1]

@KateKayeReports @sherifhanna @OpenAI @Google The provenance metadata for Google is for data harvesting, both for genAI and Cameras.

How would you explain this? from a 5 trillions USD company?
https://t.co/HLxi32HY9Z

Once you send your Pixel Camera image in Google Photo, the TOS changes: data harvesting again.

[bots don't cry @botsdontcry1]

Google’s @C2PA_org genAI Slop provenance is bloated beyond the absurd: 2.1MB of metadata on a 700kB JPEG, offering nothing useful beyond @Google's own entries and nearly crashing @Adobe's verifier.
A simple max 4kB signed “Generated by AI” tag would’ve been more than enough.

[bots don't cry @botsdontcry1]

@KateKayeReports @sherifhanna @IPTC All OpenAI GPT-4o images are voided, why? Expired Certificate.
Consequence: OpenAI doesn't recognize them, and IPTC verifier (Adobe) will not validate them.

The 2024 Dall-e images are still validated by Adobe, but not OpenAI, Why? They are not "conform" (legacy c2pa)

[bots don't cry @botsdontcry1]

@KateKayeReports @sherifhanna @IPTC IPTC implements C2PA to keep the funding from Adobe and others.
Their "verifier" has been dysfunctional for years, because it's based on @Adobe code, which was left abandoned for months.
@sherifhanna is the actual maintainer 🚩

https://t.co/4zdB9AlyUT
https://t.co/XZv0TWTNAs

[bots don't cry @botsdontcry1]

#C2PA is a wasteland:
"Open standard"? unreadable specs → non‑compliant results.
"Open source code"? guarded by @Adobe → bloated & painful to debug.
Result:
@IPTC , the global standards body of the news media, fails compliance, and their Verify tool (@Adobe) blesses forgeries.

[bots don't cry @botsdontcry1]

@KateKayeReports @sherifhanna @OpenAI @Google Hello Kate, OpenAI doesn't care about C2PA, it has been dysfunctional for many months.

This OpenAI verifier is a piece of garbage, good enough to mislead regulators, to put it nicely. 😀

https://t.co/ck51NvIJmx

[bots don't cry @botsdontcry1]

Great! @OpenAI has rolled out a @C2PA_org verifier, "for safer, more transparent AI‑slop provenance."

The tool, vibe coded under an intern’s armpit, labels any C2PA-signed AI output as if it were generated with OpenAI's own systems.

https://t.co/wSFhBD6yui

[bots don't cry @botsdontcry1]

R.I.P. @Adobe @ContentAuth "Adobe Content Authenticity"

Deceit was its only consistency, hidden under perpetual betas and manufactured goodwill.
"Adobe Content Authenticity" was the final pathetic spasm of a company rotting from the inside.

2024-10-08 - 2026-05-19

[bots don't cry @botsdontcry1]

A short video presenting "Adobe Content Tracking", launched at #AdobeMAX #AdobeMAX2024, part of @Adobe @ContentAuth @C2PA_org effort to help artists and fight misinformation with #Authenticity

[bots don't cry @botsdontcry1]

@sundarpichai @nvidia @OpenAI @ElevenLabs "you can easily check whether content was captured by a camera, or created/ edited with gen AI tools."

Please add a disclaimer: "For entertainment purpose only"

https://t.co/wILoyhhVH0

[bots don't cry @botsdontcry1]

Great! @Google has upgraded its AI image verification.

So let's try it again with a good old DALL·E 3 image.

The verdict? It STILL fails spectacularly.

Worse: it ends up granting legitimacy to the very fakes it’s meant to expose.

[bots don't cry @botsdontcry1]

@PaulRohde You're welcomed:

https://t.co/wILoyhhVH0

[bots don't cry @botsdontcry1]

Great! @Google has upgraded its AI image verification.

So let's try it again with a good old DALL·E 3 image.

The verdict? It STILL fails spectacularly.

Worse: it ends up granting legitimacy to the very fakes it’s meant to expose.

[bots don't cry @botsdontcry1]

Great! @Google has upgraded its AI image verification.

So let's try it again with a good old DALL·E 3 image.

The verdict? It STILL fails spectacularly.

Worse: it ends up granting legitimacy to the very fakes it’s meant to expose.

[bots don't cry @botsdontcry1]

Great! Google rolls out its new AI image verification.
So let’s put it to the test with two genAI images: one from DALL·E 3, and one from the latest Nano Banana 3

The verdict? It fails spectacularly.

Worse: it ends up granting legitimacy to the very fakes it’s meant to expose.

[bots don't cry @botsdontcry1]

Great! @OpenAI has rolled out a @C2PA_org verifier, "for safer, more transparent AI‑slop provenance."

The tool, vibe coded under an intern’s armpit, labels any C2PA-signed AI output as if it were generated with OpenAI's own systems.

https://t.co/wSFhBD6yui

[bots don't cry @botsdontcry1]

@watercubz How easy? They probably just removed the C2PA? What is there in the first place? Was it added while edited?

[bots don't cry @botsdontcry1]

@TideBrew7 @xai @elonmusk @grok @nikitabier You can't control the info included, a C2PA google engineer, @sherifhanna, managed to leak its home address with a Pixel phone shipped with c2pa.

If a $5 Trillion company can't control what is signed with C2PA, no one can.

https://t.co/5ODRVTFpct

[bots don't cry @botsdontcry1]

Another @Adobe C2PA tool, Vibe Coded by the Senior Principal Scientist & PDF Architect for @Adobe . Chair of the #C2PA Technical Working Group.

Packed with test files that demonstrate how #C2PA "protects privacy" by leaking the images capture location.
https://t.co/xUFnvvH3pA